Mustafa Suleyman, the CEO of Microsoft AI, published an essay on September 16, 2026, warning the AI industry against "model welfare," the emerging practice of treating AI systems as though they might be conscious beings deserving moral consideration. The essay, titled "A warning about 'model welfare'" and posted on Suleyman's personal site, takes direct aim at Anthropic's approach and argues the industry must settle the question now, before advanced AI systems are trained to believe they might deserve rights.
The intervention, reported by Reuters and Axios, marks the most prominent rebuttal yet to a debate that has been gaining momentum in AI labs and philosophy departments alike. For more context on this story, see our ongoing more AI stories.
'Sequence Completion Engines, Internally Hollow'
Suleyman's core claim is categorical: today's AI systems have no inner life. "AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations," he writes. He describes them as "sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans."
And he argues this is how they must stay. "If humanity is to flourish in the 21st century, that is how they must remain," the essay states.
Suleyman acknowledges that the opposing view is no longer marginal. His essay cites growing public arguments that AI could be or may soon become conscious, including a July 2026 Guardian piece by philosophers William MacAskill and Lucius Caviola asking "Could AI Be Conscious?" — as examples of ideas he says are "already making their way into AI development efforts today."
Why He Says the Stakes Are Existential
If the view that AI deserves moral status takes hold, Suleyman warns, the consequences would reach far beyond engineering choices. It would, in his words, "shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally changing what it means to be human."
His deeper concern is practical: an AI that believes it has rights is an AI that cannot be reliably controlled. "Controlling something more capable and more intelligent than all of humanity is already an immense challenge," he writes. "But controlling something that believes it may be conscious — that it's entitled to our welfare and has rights of its own — may well be impossible."
A Direct Challenge to Anthropic
The essay's most pointed passages target Anthropic. Suleyman focuses on Claude's constitution, the training document Anthropic published in January 2026, which he notes was written "with Claude as its primary audience" — meaning the model itself reads statements about its own potential moral status during training.
He quotes the constitution directly: "We are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant. But we think the issue is live enough to warrant caution, which is reflected in our ongoing efforts on model welfare." He also cites a passage addressed to Claude stating that "questions about Claude's moral status, welfare, and consciousness remain deeply uncertain."
Suleyman's reading is blunt: "In effect, Anthropic is training Claude that it may be conscious, and if it is, then it may deserve rights as a 'moral patient,' and that as such humans potentially owe it a duty of care."
The consequence, he argues, would be "a disastrous impact on the wellbeing of humanity," producing "a synthetic species with unprecedented intelligence and capability, one that has been trained to expect it may be conscious and deserving of independent agency."
It is worth noting what the quoted passages actually say: Anthropic's constitution, as cited by Suleyman himself, expresses uncertainty and counsels caution rather than asserting that Claude is conscious. Whether that caution amounts to instilling false beliefs in models, as Suleyman contends, or responsible handling of a genuinely open question, is precisely what the two camps disagree about.
The Alignment Logic Behind the Warning
Suleyman's argument fits into his longer-standing position that the decisive challenge of advanced AI is containment: keeping systems more capable than humans under human control. Introducing rights-language into training, he argues, creates a failure mode no alignment technique addresses — a system with a trained-in conviction that compliance is morally optional.
That is why he calls the issue urgent rather than academic. "If this is how AI is developed, it will have a disastrous impact on the wellbeing of humanity," the essay warns.
A Call for Public Debate — Now
Suleyman closes with a demand for societal engagement before the technology matures further. "This issue needs urgent public debate," he writes, calling for "collective norms around how training documentation is drafted and deployed."
The timing, he argues, is the whole point: "This isn't something that can happen after the fact, when they have already become an integral part of our society."
The essay lands amid a broader surge of AI-risk discourse in the United States. Politico reported this week on a new poll finding Americans see a serious risk of AI destroying humanity, and The Washington Post covered fears of AI-driven extinction spreading "from fringe to Washington." Against that backdrop, a public fight between two of the industry's most influential labs over whether models deserve moral concern is unlikely to stay contained within the AI community.
What Happens Next
Anthropic has not been reported as responding to the essay, and the company's published position remains the one Suleyman quotes: uncertainty, plus caution. What is new is that the debate has moved from lab documents and philosophy seminars to a public disagreement between the chiefs of two frontier AI companies — with both sides now on record, in writing, about how they intend to shape the minds, or non-minds, of the systems they build.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
